Error detection enhanced decoding of difference set codes for memory applications
Download as PPTX, PDF0 likes1,845 views
This document discusses error detection techniques for memory applications using difference set codes. It begins by introducing difference set codes and the (21,11) difference set code. It then describes the conventional decoder for this code and issues with silent data corruption when decoding words with 3 or more errors. The document proposes an error detection majority logic decoding technique that can detect errors in 3 cycles to reduce latency and detect uncorrectable errors. This approach avoids the issues with the conventional decoder while enhancing error detection capabilities for memory applications.
![30
References
[1]S. Liu, P. Reviriego and J. A. Maestro “Error-Detection Enhanced Decoding
of Difference Set Codes for Memory Applications ” IEEE Trans. on Device
and Materials Reliability, Vol. 12, No. 2, 2012, pp 335 – 340.
[2]S. Liu, P. Reviriego and J. A. Maestro “Efficient Majority Logic Fault
Detection with Difference-Set Codes for Memory Applications” IEEE Trans.
on Very Large Scale Integration Systems, Vol. 20, No. 1, 2012, pp 148-156.
[3]S. Lin and D. J. Costello, Error Control Coding, 2nd ed. Englewood
Cliffs, NJ: Prentice-Hall, 2004.](https://image.slidesharecdn.com/errordetectionenhanceddecodingofdifferencesetcodesformemoryapplications-120927022519-phpapp01/85/Error-detection-enhanced-decoding-of-difference-set-codes-for-memory-applications-30-320.jpg)
Error detection enhanced decoding of difference set codes for memory applications
- 1. 1 Error Detection Enhanced Decoding of Difference Set Codes for Memory Applications SHERIN DEENA SAM
- 2. 2 What a difference a bit makes!!! • NASA engineers believe they have traced the cause of Voyager 2's glitch (in 2010) to single flip of bit in the spacecraft’s memory. • "A value in a single memory location was changed from a 0 to a 1," said JPL’s Veronia McGregor.
- 3. 3 Contents • Error Correcting Code Memory • OSMLD Difference Set Codes (21,11) DS code • Conventional Decoder for (21,11) DS code Avoiding silent data corruption • Error Detection Majority Logic Decoding
- 4. 4 Error Correcting Code Memory
- 5. 5 ECC memory • Error correcting code memory is a type of computer data storage that can detect and correct common kinds of internal data corruption. • It is used in computers where data corruption cannot be tolerated under any circumstances, such as for scientific or financial computing, as spacecrafts and as servers. • Electrical or magnetic interference can cause a single bit to spontaneously flip to the opposite state. • Majority of soft errors in RAM chips occur as a result of ionizing radiation. • ECC is used in NAND flash, SRAM, DRAM, SDRAM etc. • ECC schemes used are BCH Error Correction Code, Euclidean Geometry ECC, Difference Set ECC.
- 6. 6 OSMLD Difference Set Codes DS codes are attractive because they are One Step Majority Logic Decodable (OSMLD).
- 7. 7 Difference Set • P={l0,l1,l2,……,lq} is a set with q+1 nonnegative integers such that 0≤lo<l1<l2<…..<lq ≤q(q+1) • q(q+1) ordered differences D={lj-li:j≠i} • P is a perfect simple difference-set of order q iff 1. All positive differences in D are distinct. 2. All negative differences in D are distinct. 3. If lj-li is a negative difference in D, then q(q+1)+1+(lj-li) is not equal to any positive difference in D. • Consider the set P= {0,2,7,8,11} with q=4 • 4.5=20 ordered differences D={2,7,8,11,5,6,9,1,4,3, -2,-7,-8,-11,-5,-6,-9,-1,-4,-3}
- 8. 8 (21,11) DS Cyclic Code • We define a polynomial • n=q(q+1)+1=21 • Parity-check polynomial • k=degree of h(x)=11 • Generator polynomial
- 9. 9 Code Parameters • Code length=n=q(q+1)+1=21 • Number of parity check digits=n-k=10 • Number of correctable errors=t=2 • Number of detectable errors=t+s=3 • Number of parity check equations=2t+1=5 • Minimum distance=d=q+2=6=2t+s+1 • So, this is a Double Error Correction Triple Error Detection (DEC- TED) Code
- 10. 10 Dual Code • (n, n-k) cyclic code. • Code generated by • It gives the dual code of the code generated by g(x). • It is in the null space of the DS code generated by g(x). • We define a polynomial
- 11. 11 • We define • Shifting w0(x) cyclically to the right 2 times, 7 times, 8 times, and 11 times, we obtain • w0(x),w1(x), w2(x), w3(x), w4(x) are 5 polynomials orthogonal on
- 12. 12 • We can form 5 parity-check sums orthogonal on e20. • A1=e9+e12+e13+ e18+e20 • A2=e1+e11+e14+e15+e20 • A3=e4+e6+e16+e19+e20 • A4=e0+e5+e7+e17+e20 • A5=e2+e3+e8+e10+e20 • e20 is checked by all 5 check sums and no other error digit is checked by more than one check sum.
- 13. 13 One Step Majority Logic Decoding • If e20 = 1 and if there is one or no error occurring among the other 20 digit positions, then at least 4 of the 5 check sums are equal to 1. • If e20= 0 and if there are two or fewer errors occurring among the other 20 digit positions, then at least 3 of the 5 check sums are equal to 0. ▫ If there are two or fewer errors , the one step majority logic decoding always results in correct decoding of e20. • Consider e9= e1= e4= 1. We have A1=1, A2=1, A3=1, A4=0,A5=0. Since the majority of the five sums is 1, e20 is decoded as 1. This results in an incorrect decoding. • Thus, by one-step majority logic decoding, the code is capable of correcting any error pattern with two or fewer errors.
- 14. 14 Conventional Decoder for (21,11) DS code
- 15. 15 Decoder for (21,11) DS code
- 16. 16 Correction of 2 errors
- 17. 17 SDC- Silent Data Corruption • Considering the (21,11) DS code, when a triple error occurs, it can cause Silent Data Corruption (SDC) because it can trigger a miscorrection that then can result in the word being decoded into another valid coded word.
- 18. 18 SDC on decoding word with 3 errors
- 19. 19 Avoiding SDC • To avoid SDC, the proposed decoder for the (21,11) code is the same but requiring at least four check equations to be one to perform a bit correction. • Requiring at least t+2 ones instead of a majority of ones will ensure that there is no miscorrection with t+1 errors and at the same time guarantees that t errors are corrected.
- 20. 20 Avoiding SDC with 3 errors
- 21. 21 Issues • The decoder requires 21 iterations to decode a word. This results in a large latency that in a memory application would increase the access time. • Additionally, the number of iterations would increase linearly with the length of the coded word.
- 22. 22 Error Detection Majority Logic Decoding (EDMLD)
- 23. 23 Avoiding decoding latency • All errors affecting five or less bits are detected in the first three iterations. • This can be used to reduce the decoding latency by stopping the process when no error has been detected in the first three iterations. • This technique reduces the decoding latency significantly as most words will have no errors.
- 24. 24 No error detected in 3 cycles
- 25. 25 Error detected in 3 cycles
- 26. 26 Detection of uncorrectable errors • Additionally, it can also be used to detect uncorrected errors once the decoding has been completed. • This will be useful to detect errors that affect t+1 bits and can not be corrected.
- 29. 29 Conclusion • The decoder has been implemented in Verilog HDL. • The conventional decoder has been modified to avoid SDC. • Error Detection Majority Logic Decoding (EDMLD) has been done which provides detection of errors in addition to correction. ▫ It reduces latency. ▫ It detects errors that affect t+1 bits and can not be corrected.
- 30. 30 References [1]S. Liu, P. Reviriego and J. A. Maestro “Error-Detection Enhanced Decoding of Difference Set Codes for Memory Applications ” IEEE Trans. on Device and Materials Reliability, Vol. 12, No. 2, 2012, pp 335 – 340. [2]S. Liu, P. Reviriego and J. A. Maestro “Efficient Majority Logic Fault Detection with Difference-Set Codes for Memory Applications” IEEE Trans. on Very Large Scale Integration Systems, Vol. 20, No. 1, 2012, pp 148-156. [3]S. Lin and D. J. Costello, Error Control Coding, 2nd ed. Englewood Cliffs, NJ: Prentice-Hall, 2004.
- 31. 31 THANK YOU